@vellumai/assistant 0.10.0 → 0.10.1-staging.1

package/src/memory/migrations/140-backfill-usage-cache-accounting.ts

@@ -6,7 +6,6 @@ import type {
 import { getLogger } from "../../util/logger.js";
 import { resolvePricingForUsageWithOverrides } from "../../util/pricing.js";
 import { type DrizzleDb, getSqliteFrom } from "../db-connection.js";
-import { withCrashRecovery } from "./validate-migration-state.js";
 
 /**
  * Reverse v20: no-op — cannot reliably identify which llm_usage_events rows
@@ -23,8 +22,6 @@ export function downBackfillUsageCacheAccounting(_database: DrizzleDb): void {
 
 const log = getLogger("memory-db");
 
-const CHECKPOINT_KEY = "migration_backfill_usage_cache_accounting_v1";
-
 interface UsageEventRow {
   id: string;
   conversation_id: string | null;
@@ -145,226 +142,224 @@ function buildRequestLogMap(
 }
 
 export function migrateBackfillUsageCacheAccounting(database: DrizzleDb): void {
-  withCrashRecovery(database, CHECKPOINT_KEY, () => {
-    const raw = getSqliteFrom(database);
-
-    const usageEventsTableExists = raw
-      .query(
-        `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'llm_usage_events'`,
-      )
-      .get();
-    const requestLogsTableExists = raw
-      .query(
-        `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'llm_request_logs'`,
-      )
-      .get();
-    if (!usageEventsTableExists || !requestLogsTableExists) return;
-
-    const usageRows = raw
-      .query(
-        /*sql*/ `
-        SELECT
-          id,
-          conversation_id,
-          created_at,
-          provider,
-          model,
-          input_tokens,
-          output_tokens,
-          cache_creation_input_tokens,
-          cache_read_input_tokens
-        FROM llm_usage_events
-        ORDER BY conversation_id ASC, created_at ASC, id ASC
-        `,
-      )
-      .all() as UsageEventRow[];
-
-    const requestLogRows = raw
-      .query(
-        /*sql*/ `
-        SELECT
-          id,
-          conversation_id,
-          response_payload,
-          created_at
-        FROM llm_request_logs
-        ORDER BY conversation_id ASC, created_at ASC, id ASC
-        `,
-      )
-      .all() as RequestLogRow[];
-
-    const requestLogsByConversation = buildRequestLogMap(requestLogRows);
-    const requestOffsets = new Map<string, number>();
-    const previousUsageEventCreatedAt = new Map<string, number>();
-    const pricingOverrides = getConfig().llm.pricingOverrides;
-
-    let scannedAnthropicRows = 0;
-    let updatedRows = 0;
-    let skippedNoConversation = 0;
-    let skippedNoLogs = 0;
-    let ignoredUnusableLogs = 0;
-    let skippedInconsistentRows = 0;
-
-    const updateRow = raw.prepare(/*sql*/ `
-      UPDATE llm_usage_events
-      SET
-        input_tokens = ?,
-        output_tokens = ?,
-        cache_creation_input_tokens = ?,
-        cache_read_input_tokens = ?,
-        estimated_cost_usd = ?,
-        pricing_status = ?
-      WHERE id = ?
-      `);
-
-    try {
-      raw.exec("BEGIN");
-
-      for (const usageRow of usageRows) {
-        const conversationId = usageRow.conversation_id;
-        if (conversationId == null) {
-          if (usageRow.provider === "anthropic") {
-            scannedAnthropicRows += 1;
-            skippedNoConversation += 1;
-          }
-          continue;
-        }
+  const raw = getSqliteFrom(database);
+
+  const usageEventsTableExists = raw
+    .query(
+      `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'llm_usage_events'`,
+    )
+    .get();
+  const requestLogsTableExists = raw
+    .query(
+      `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'llm_request_logs'`,
+    )
+    .get();
+  if (!usageEventsTableExists || !requestLogsTableExists) return;
+
+  const usageRows = raw
+    .query(
+      /*sql*/ `
+      SELECT
+        id,
+        conversation_id,
+        created_at,
+        provider,
+        model,
+        input_tokens,
+        output_tokens,
+        cache_creation_input_tokens,
+        cache_read_input_tokens
+      FROM llm_usage_events
+      ORDER BY conversation_id ASC, created_at ASC, id ASC
+      `,
+    )
+    .all() as UsageEventRow[];
+
+  const requestLogRows = raw
+    .query(
+      /*sql*/ `
+      SELECT
+        id,
+        conversation_id,
+        response_payload,
+        created_at
+      FROM llm_request_logs
+      ORDER BY conversation_id ASC, created_at ASC, id ASC
+      `,
+    )
+    .all() as RequestLogRow[];
+
+  const requestLogsByConversation = buildRequestLogMap(requestLogRows);
+  const requestOffsets = new Map<string, number>();
+  const previousUsageEventCreatedAt = new Map<string, number>();
+  const pricingOverrides = getConfig().llm.pricingOverrides;
+
+  let scannedAnthropicRows = 0;
+  let updatedRows = 0;
+  let skippedNoConversation = 0;
+  let skippedNoLogs = 0;
+  let ignoredUnusableLogs = 0;
+  let skippedInconsistentRows = 0;
+
+  const updateRow = raw.prepare(/*sql*/ `
+    UPDATE llm_usage_events
+    SET
+      input_tokens = ?,
+      output_tokens = ?,
+      cache_creation_input_tokens = ?,
+      cache_read_input_tokens = ?,
+      estimated_cost_usd = ?,
+      pricing_status = ?
+    WHERE id = ?
+    `);
 
-        const previousCreatedAt =
-          previousUsageEventCreatedAt.get(conversationId) ?? null;
-        const conversationRequestLogs =
-          requestLogsByConversation.get(conversationId) ?? [];
-        let requestOffset = requestOffsets.get(conversationId) ?? 0;
-        const windowLogs: RequestLogRow[] = [];
-
-        while (
-          requestOffset < conversationRequestLogs.length &&
-          conversationRequestLogs[requestOffset]!.created_at <=
-            usageRow.created_at
-        ) {
-          const requestLog = conversationRequestLogs[requestOffset]!;
-          if (
-            previousCreatedAt == null ||
-            requestLog.created_at > previousCreatedAt
-          ) {
-            windowLogs.push(requestLog);
-          }
-          requestOffset += 1;
+  try {
+    raw.exec("BEGIN");
+
+    for (const usageRow of usageRows) {
+      const conversationId = usageRow.conversation_id;
+      if (conversationId == null) {
+        if (usageRow.provider === "anthropic") {
+          scannedAnthropicRows += 1;
+          skippedNoConversation += 1;
         }
+        continue;
+      }
 
-        requestOffsets.set(conversationId, requestOffset);
-        previousUsageEventCreatedAt.set(conversationId, usageRow.created_at);
-
-        if (usageRow.provider !== "anthropic") continue;
-
-        scannedAnthropicRows += 1;
-        if (windowLogs.length === 0) {
-          skippedNoLogs += 1;
-          continue;
+      const previousCreatedAt =
+        previousUsageEventCreatedAt.get(conversationId) ?? null;
+      const conversationRequestLogs =
+        requestLogsByConversation.get(conversationId) ?? [];
+      let requestOffset = requestOffsets.get(conversationId) ?? 0;
+      const windowLogs: RequestLogRow[] = [];
+
+      while (
+        requestOffset < conversationRequestLogs.length &&
+        conversationRequestLogs[requestOffset]!.created_at <=
+          usageRow.created_at
+      ) {
+        const requestLog = conversationRequestLogs[requestOffset]!;
+        if (
+          previousCreatedAt == null ||
+          requestLog.created_at > previousCreatedAt
+        ) {
+          windowLogs.push(requestLog);
         }
+        requestOffset += 1;
+      }
 
-        let directInputTokens = 0;
-        let outputTokens = 0;
-        let cacheReadInputTokens = 0;
-        let cacheCreationInputTokens = 0;
-        let cacheCreation5mInputTokens = 0;
-        let cacheCreation1hInputTokens = 0;
-        let usableLogCount = 0;
-
-        for (const requestLog of windowLogs) {
-          const reconstructedUsage = parseResponseUsage(
-            requestLog.response_payload,
-          );
-          if (!reconstructedUsage) {
-            ignoredUnusableLogs += 1;
-            continue;
-          }
-          usableLogCount += 1;
-          directInputTokens += reconstructedUsage.directInputTokens;
-          outputTokens += reconstructedUsage.outputTokens;
-          cacheReadInputTokens += reconstructedUsage.cacheReadInputTokens;
-          cacheCreationInputTokens +=
-            reconstructedUsage.cacheCreationInputTokens;
-          cacheCreation5mInputTokens +=
-            reconstructedUsage.anthropicCacheCreation
-              ?.ephemeral_5m_input_tokens ?? 0;
-          cacheCreation1hInputTokens +=
-            reconstructedUsage.anthropicCacheCreation
-              ?.ephemeral_1h_input_tokens ?? 0;
-        }
+      requestOffsets.set(conversationId, requestOffset);
+      previousUsageEventCreatedAt.set(conversationId, usageRow.created_at);
 
-        if (usableLogCount === 0) {
-          skippedNoLogs += 1;
-          continue;
-        }
+      if (usageRow.provider !== "anthropic") continue;
 
-        const totalInputTokens =
-          directInputTokens + cacheCreationInputTokens + cacheReadInputTokens;
-        const existingTotalInputTokens =
-          Math.max(usageRow.input_tokens, 0) +
-          Math.max(usageRow.cache_creation_input_tokens ?? 0, 0) +
-          Math.max(usageRow.cache_read_input_tokens ?? 0, 0);
+      scannedAnthropicRows += 1;
+      if (windowLogs.length === 0) {
+        skippedNoLogs += 1;
+        continue;
+      }
 
-        // This lets the migration safely rewrite both flattened historical rows
-        // and already-correct rows, while skipping mismatched request-log windows.
-        if (
-          totalInputTokens < 0 ||
-          directInputTokens < 0 ||
-          totalInputTokens !== existingTotalInputTokens ||
-          outputTokens !== Math.max(usageRow.output_tokens, 0)
-        ) {
-          skippedInconsistentRows += 1;
+      let directInputTokens = 0;
+      let outputTokens = 0;
+      let cacheReadInputTokens = 0;
+      let cacheCreationInputTokens = 0;
+      let cacheCreation5mInputTokens = 0;
+      let cacheCreation1hInputTokens = 0;
+      let usableLogCount = 0;
+
+      for (const requestLog of windowLogs) {
+        const reconstructedUsage = parseResponseUsage(
+          requestLog.response_payload,
+        );
+        if (!reconstructedUsage) {
+          ignoredUnusableLogs += 1;
           continue;
         }
+        usableLogCount += 1;
+        directInputTokens += reconstructedUsage.directInputTokens;
+        outputTokens += reconstructedUsage.outputTokens;
+        cacheReadInputTokens += reconstructedUsage.cacheReadInputTokens;
+        cacheCreationInputTokens +=
+          reconstructedUsage.cacheCreationInputTokens;
+        cacheCreation5mInputTokens +=
+          reconstructedUsage.anthropicCacheCreation
+            ?.ephemeral_5m_input_tokens ?? 0;
+        cacheCreation1hInputTokens +=
+          reconstructedUsage.anthropicCacheCreation
+            ?.ephemeral_1h_input_tokens ?? 0;
+      }
 
-        const pricingUsage: PricingUsage = {
-          directInputTokens,
-          outputTokens,
-          cacheCreationInputTokens,
-          cacheReadInputTokens,
-          anthropicCacheCreation:
-            cacheCreationInputTokens > 0
-              ? {
-                  ephemeral_5m_input_tokens: cacheCreation5mInputTokens,
-                  ephemeral_1h_input_tokens: cacheCreation1hInputTokens,
-                }
-              : null,
-        };
-        const pricing = resolvePricingForUsageWithOverrides(
-          usageRow.provider,
-          usageRow.model,
-          pricingUsage,
-          pricingOverrides,
-        );
+      if (usableLogCount === 0) {
+        skippedNoLogs += 1;
+        continue;
+      }
 
-        updateRow.run(
-          directInputTokens,
-          outputTokens,
-          cacheCreationInputTokens,
-          cacheReadInputTokens,
-          pricing.estimatedCostUsd,
-          pricing.pricingStatus,
-          usageRow.id,
-        );
-        updatedRows += 1;
+      const totalInputTokens =
+        directInputTokens + cacheCreationInputTokens + cacheReadInputTokens;
+      const existingTotalInputTokens =
+        Math.max(usageRow.input_tokens, 0) +
+        Math.max(usageRow.cache_creation_input_tokens ?? 0, 0) +
+        Math.max(usageRow.cache_read_input_tokens ?? 0, 0);
+
+      // This lets the migration safely rewrite both flattened historical rows
+      // and already-correct rows, while skipping mismatched request-log windows.
+      if (
+        totalInputTokens < 0 ||
+        directInputTokens < 0 ||
+        totalInputTokens !== existingTotalInputTokens ||
+        outputTokens !== Math.max(usageRow.output_tokens, 0)
+      ) {
+        skippedInconsistentRows += 1;
+        continue;
       }
 
-      raw.exec("COMMIT");
-    } catch (err) {
-      raw.exec("ROLLBACK");
-      throw err;
+      const pricingUsage: PricingUsage = {
+        directInputTokens,
+        outputTokens,
+        cacheCreationInputTokens,
+        cacheReadInputTokens,
+        anthropicCacheCreation:
+          cacheCreationInputTokens > 0
+            ? {
+                ephemeral_5m_input_tokens: cacheCreation5mInputTokens,
+                ephemeral_1h_input_tokens: cacheCreation1hInputTokens,
+              }
+            : null,
+      };
+      const pricing = resolvePricingForUsageWithOverrides(
+        usageRow.provider,
+        usageRow.model,
+        pricingUsage,
+        pricingOverrides,
+      );
+
+      updateRow.run(
+        directInputTokens,
+        outputTokens,
+        cacheCreationInputTokens,
+        cacheReadInputTokens,
+        pricing.estimatedCostUsd,
+        pricing.pricingStatus,
+        usageRow.id,
+      );
+      updatedRows += 1;
     }
 
-    log.info(
-      {
-        scannedAnthropicRows,
-        updatedRows,
-        skippedNoConversation,
-        skippedNoLogs,
-        ignoredUnusableLogs,
-        skippedInconsistentRows,
-      },
-      "Backfilled historical Anthropic usage rows from request-log accounting",
-    );
-  });
+    raw.exec("COMMIT");
+  } catch (err) {
+    raw.exec("ROLLBACK");
+    throw err;
+  }
+
+  log.info(
+    {
+      scannedAnthropicRows,
+      updatedRows,
+      skippedNoConversation,
+      skippedNoLogs,
+      ignoredUnusableLogs,
+      skippedInconsistentRows,
+    },
+    "Backfilled historical Anthropic usage rows from request-log accounting",
+  );
 }

package/src/memory/migrations/141-rename-verification-table.ts

@@ -1,5 +1,4 @@
 import { type DrizzleDb, getSqliteFrom } from "../db-connection.js";
-import { withCrashRecovery } from "./validate-migration-state.js";
 
 /**
  * Reverse v21: rename channel_verification_sessions back to
@@ -61,57 +60,55 @@ export function downRenameVerificationTable(database: DrizzleDb): void {
  * old table name.
  */
 export function migrateRenameVerificationTable(database: DrizzleDb): void {
-  withCrashRecovery(database, "migration_rename_verification_table_v1", () => {
-    const raw = getSqliteFrom(database);
+  const raw = getSqliteFrom(database);
 
-    // Check the old table exists before attempting anything
-    const oldTableExists = raw
-      .query(
-        `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'channel_guardian_verification_challenges'`,
-      )
-      .get();
-    if (!oldTableExists) return;
+  // Check the old table exists before attempting anything
+  const oldTableExists = raw
+    .query(
+      `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'channel_guardian_verification_challenges'`,
+    )
+    .get();
+  if (!oldTableExists) return;
 
-    // If the new table already exists, the rename would collide — skip
-    const newTableExists = raw
-      .query(
-        `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'channel_verification_sessions'`,
-      )
-      .get();
-    if (newTableExists) return;
+  // If the new table already exists, the rename would collide — skip
+  const newTableExists = raw
+    .query(
+      `SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = 'channel_verification_sessions'`,
+    )
+    .get();
+  if (newTableExists) return;
 
-    // Rename the physical table
-    raw.exec(
-      /*sql*/ `ALTER TABLE channel_guardian_verification_challenges RENAME TO channel_verification_sessions`,
-    );
+  // Rename the physical table
+  raw.exec(
+    /*sql*/ `ALTER TABLE channel_guardian_verification_challenges RENAME TO channel_verification_sessions`,
+  );
 
-    // Drop and recreate indexes that referenced the old table name.
-    // SQLite automatically updates index table references on RENAME, but the
-    // index names still reference the old naming convention — drop and recreate
-    // with consistent names pointing at the new table.
+  // Drop and recreate indexes that referenced the old table name.
+  // SQLite automatically updates index table references on RENAME, but the
+  // index names still reference the old naming convention — drop and recreate
+  // with consistent names pointing at the new table.
 
-    raw.exec(
-      /*sql*/ `DROP INDEX IF EXISTS idx_channel_guardian_challenges_lookup`,
-    );
-    raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_active`);
-    raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_identity`);
-    raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_destination`);
-    raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_bootstrap`);
+  raw.exec(
+    /*sql*/ `DROP INDEX IF EXISTS idx_channel_guardian_challenges_lookup`,
+  );
+  raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_active`);
+  raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_identity`);
+  raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_destination`);
+  raw.exec(/*sql*/ `DROP INDEX IF EXISTS idx_guardian_sessions_bootstrap`);
 
-    raw.exec(
-      /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_lookup ON channel_verification_sessions(channel, challenge_hash, status)`,
-    );
-    raw.exec(
-      /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_active ON channel_verification_sessions(channel, status)`,
-    );
-    raw.exec(
-      /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_identity ON channel_verification_sessions(channel, expected_external_user_id, expected_chat_id, status)`,
-    );
-    raw.exec(
-      /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_destination ON channel_verification_sessions(channel, destination_address)`,
-    );
-    raw.exec(
-      /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_bootstrap ON channel_verification_sessions(channel, bootstrap_token_hash, status)`,
-    );
-  });
+  raw.exec(
+    /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_lookup ON channel_verification_sessions(channel, challenge_hash, status)`,
+  );
+  raw.exec(
+    /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_active ON channel_verification_sessions(channel, status)`,
+  );
+  raw.exec(
+    /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_identity ON channel_verification_sessions(channel, expected_external_user_id, expected_chat_id, status)`,
+  );
+  raw.exec(
+    /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_destination ON channel_verification_sessions(channel, destination_address)`,
+  );
+  raw.exec(
+    /*sql*/ `CREATE INDEX IF NOT EXISTS idx_verification_sessions_bootstrap ON channel_verification_sessions(channel, bootstrap_token_hash, status)`,
+  );
 }

package/src/memory/migrations/142-rename-verification-session-id-column.ts

@@ -1,5 +1,4 @@
 import { type DrizzleDb, getSqliteFrom } from "../db-connection.js";
-import { withCrashRecovery } from "./validate-migration-state.js";
 
 /**
  * Reverse v22: rename verification_session_id back to
@@ -34,29 +33,23 @@ export function downRenameVerificationSessionIdColumn(
 export function migrateRenameVerificationSessionIdColumn(
   database: DrizzleDb,
 ): void {
-  withCrashRecovery(
-    database,
-    "migration_rename_verification_session_id_column_v1",
-    () => {
-      const raw = getSqliteFrom(database);
+  const raw = getSqliteFrom(database);
 
-      // Check the old column exists and the new column doesn't before attempting the rename.
-      // Both checks are needed for crash recovery: if the rename succeeded but the checkpoint
-      // didn't commit, the old column is gone and the new one already exists.
-      const columns = raw
-        .query(`PRAGMA table_info(call_sessions)`)
-        .all() as Array<{ name: string }>;
-      const hasOldColumn = columns.some(
-        (c) => c.name === "guardian_verification_session_id",
-      );
-      const hasNewColumn = columns.some(
-        (c) => c.name === "verification_session_id",
-      );
-      if (!hasOldColumn || hasNewColumn) return;
+  // Check the old column exists and the new column doesn't before attempting the rename.
+  // Both checks are needed for crash recovery: if the rename succeeded but the checkpoint
+  // didn't commit, the old column is gone and the new one already exists.
+  const columns = raw
+    .query(`PRAGMA table_info(call_sessions)`)
+    .all() as Array<{ name: string }>;
+  const hasOldColumn = columns.some(
+    (c) => c.name === "guardian_verification_session_id",
+  );
+  const hasNewColumn = columns.some(
+    (c) => c.name === "verification_session_id",
+  );
+  if (!hasOldColumn || hasNewColumn) return;
 
-      raw.exec(
-        /*sql*/ `ALTER TABLE call_sessions RENAME COLUMN guardian_verification_session_id TO verification_session_id`,
-      );
-    },
+  raw.exec(
+    /*sql*/ `ALTER TABLE call_sessions RENAME COLUMN guardian_verification_session_id TO verification_session_id`,
   );
 }